is Gemini realtime model (live api) remotely useable for anyone? The gemini ai studio experiences seem to work smoothly, but i've never had luck with it bootstrapped on livekit agents sdk in full or half pipe. Either it self-interrupts or fails to support consistent spec. OpenAI has been pretty turnkey.

Not really. I can get it about 80% of the way there but not that last 20. Still getting occasional interrupt/repeats